Scientist Stories: Shawn Burgess, Translational Genome Editing

Hello everyone, and welcome to the first NHGRI genome short course completely done virtually. My name is Sean Burgess. Hopefully we can get through this without the dogs barking. But I'm going to talk today about a topic that's been in the forefront of the news for the last few years. And that's genome editing. And the reason genome editing is, uh, so important to us is that it covers a lot of areas of both human health and welfare. This is from a review about 6 years ago now where the various aspects that gene editing touches on are Things like drug development, developing animal models to study human disease, for understanding genetic variation, biomaterials becomes important in doing biotechnology and other organisms, making more robust foods, of course, biofuels. And ultimately the goal, of course, for many of us is can you do surgery on your genome and repair genetic diseases. So there's a lot of things that gene editing can touch upon. It turns out that humans have been doing gene editing even before they knew what a gene was. And if you look at the timeline, there was genetics being done through spontaneous mutations. Oh, grow over 5,000 years ago, we started selecting for dogs with certain behaviors and cows that were more robust and gave more beef or milk. Eventually, scientists learned that you can induce mutations and then select for specific phenotypes. They did it through radiation, through chemicals, and through DNA insertion of mobile elements. And then that was the sort of birth of modern genetics, and that was about 100 years ago. About 35 years ago, scientists figured out a technique called homologous recombination where they could take large pieces of DNA that matched the genome of a specific organism and they could insert sequences between those two arms and then the DNA would find the right place in the genome and swap out the old sequence and swap in the new one. It didn't work in every organism, but it did work in 3 primarily: bacteria, yeast, and mice. And that sort of created an explosion of research in those particular model organisms. About 19 years ago, the very first instance of what we're going to talk about today, this really, this rise in the new form of gene editing was developed. And these are these targeted nucleases. The first ones that came out were the ones called zinc finger nucleases, and that was about 19 years ago. And then there was a shift not too long after that to ones called transcriptional activator-like effector nucleases, or TALENs. That was 12 years ago. And then, unless you've been living under a rock for the last few years, the most exciting and revolutionary version of that was from Clustered Regularly Interspaced Short Palindromic Repeats, more commonly known as CRISPR or CRISPR-Cas9. And that really hit the news about 8 years ago. I'll show you the research that's been going on for longer than that. So I'm going to talk about these three targeted mutagenesis approaches very briefly with zinc fingers and TALENs because they've really been overwhelmed with CRISPR-Cas. But there are pros and cons to each technique and I'll talk a little bit about them. But the zinc fingers, as I said, were the first. The number of targets in the genome is fairly limited and they're, they're, they're are tough to make, so making a lot of them is— the throughput is relatively low. TALENs are probably the most flexible design. They can literally target anywhere in the genome, but they're still a little harder to make than CRISPR-Cas targets are. And then of course the CRISPRs, they're cheap, they're simple, and they're incredibly robust. There are some sequence constraints but the ability to generate enormous number of targets very easily is what makes them so exciting. So let's start with the zinc finger nucleases. And this technology evolved from the study of transcription factors. And I'm sure you know that a transcription factor is a protein that can bind to DNA, and it binds to DNA in a sequence-specific manner. So it scans the genome for a specific sequence, usually somewhere in the order of 4 to 6 bases long. It binds to the DNA and then it has an effect on the local gene expression, either turning it on or turning it off. But these are, there's over 1,500 different transcription factors in your average eukaryotic, or at least vertebrate organism. And one of the most well studied of the transcription factors, even at the atomic level, are ones called zinc fingers. And they're given the name zinc fingers because there's a sequence that's always in these transcription factors and it involves histidines and cysteines and they coordinate that green dot in the middle, which is the zinc molecule. And if you squint enough and twist your head sideways, they think that this protein structure looks like a finger, so they call it a zinc finger. And the trick about these zinc fingers is if you look at that spiral down at the bottom, that part is the part that always binds to DNA. And it turns out after studying several hundred different versions of zinc fingers and what they prefer to bind to, you can develop some rules about how, what amino acids are required to bind to what sequences. The problem is these rules are complicated. They don't always follow the rules, and it's a little difficult to predict, but it is possible, at least in theory, to link multiple zinc fingers together and create a binding site with between 3 and 4 zinc fingers that will recognize a 9 to 12 base pair sequence in the genome relatively specifically. So that became an interesting idea is if you can control where a protein binds in the genome, can you use that to manipulate the genome? And so some clever scientists went through and they took these zinc finger proteins that can now find specific sites in the genome and they linked it to a nuclease. And this nuclease has been broken in half. So you can see on this, there's a FokI and an FogI domain. And when those two are brought spatially together, they work together to cut the DNA and it may create a double-stranded break. So now if you have FokI linked to a 9-base pair recognizing sequence and the FogI linked to another 9-base pair sequence, you can theoretically find an 18-base site in the genome and cut only that site specifically. 18 bases is usually enough to find a unique site in, say, the human genome. And so this is a way to target very specifically a cut, single cut in the genome. Well, that was exciting. Like I said, they were pretty hard to make. And other researchers were studying these bacterial proteins called TALEs or TALE effectors. And they too were, were transcription factors. They're transcription factors that bacteria use to drive gene expression in their hosts. And it turns out that the rules for identifying genome sequence in the tal effectors is actually much, much simpler than they are for the zinc finger nucleases. And so the way they work is there are these repeats and they're 34 amino acids long. And within these repeats, there are just 2 bases, and they're marked in red. I don't know if you can see right here. And depending on what those 2 bases are, that determines which DNA nucleotide is bound by each of the repeat modules. So it's a simple replacement, and here's the code along the bottom here. Whichever one of those in the repeat determines your sequence. So if you want a 12-base recognition sequence, you put in 12 repeats and with those specific nucleotides swapped out to determine the DNA binding domain. And they worked on modular construction of these, so it's an easy multi-step ligation process into a backbone. And these can be done relatively quickly, although it does take some work in the lab to produce one. I would say an average lab can generate one of these zinc fingers in a week or two. High-throughput labs, it could take 2 or 3 days. Now you do the exact same strategy that we did with the zinc finger nucleases. You take TALEN-A and TALEN-B, and they have 9 to 12 base sequence specificities each. You find a location in the genome with 18 to 24 base sequence, and you get a double-stranded break right at the site. You've targeted. Like I said, starting about 8 years ago, there was an enormous wave of research linked to the CRISPR system, and I'm going to spend most of my time talking about this because it's incredibly powerful and it's generated an enormous amount of excitement in the scientific community, even in the sort of lay scientific community. And you can see it's been tested in many, many different systems and genomes of bacteria, humans, animals, plants, and everywhere it's tested, it seems to work. So how did we discover these? It actually started about 1987 when a Japanese group was sequencing the genome of a bacteria and they saw this weird a cluster of repeated sequences that they didn't understand, and it took a long time for people to actually understand what they were, even though they started seeing them more and more. So almost 15, 20 years later, when there were a lot of genomes being sequenced, bacterial genomes being sequenced, they started to realize and and phage sequences, sorry, they started to realize that these repeat sequences appeared to be coming from other organisms and were being incorporated into the genome specifically. And then eventually in 2007, they realized that these repeat sequences were a form of immunity that bacteria were developing to protect themselves against essentially viral infections. And it took a few more years for them to figure out that these repeats were working in coordination with a number of genes called Cas genes. And these genes were making RNA from the repeats and then using those RNA to target invasive DNA and cut it. So essentially you were getting an RNA-driven sequence recognition protein that would then cut DNA, do double-stranded breaks in protein— sorry, in DNA. And it turns out there are a lot of different strategies because there are literally billions of different kinds of bacteria out there. But the most popular version of this protective strategy were the ones called the Cas9. And the reason it was— the Cas9s are popular among scientists is because it only takes the one protein, Cas9, plus the RNA to do a targeted DNA cut. Some of the other versions use multiple proteins as a complex. And you can, you can split out the different bacterial species based on the particular strategy of immunity that they use. So the way this works is there's the Cas gene cluster, and then there's that CRISPR array. And then there, as I said, there's 3 types. There's Type 1, Type 2, Type 3. The Type 1 has a cascade complex using multiple proteins, same with the Type 3, but the Type 2s use this one Cas9. So the CRISPR arrays generate these RNAs. These RNAs get chopped up into short pieces, and then the Cas9 binds the RNA. And within the RNA, there is a a stretch— whoops— a stretch of RNA that matches DNA of the invading species. And when those two things— when the RNA matches the DNA, the Cas9 will cut right at that spot. So how do you get Cas9 to work? So you need to express the protein. And so because they're bacterial proteins, if you want it to work in a, in a species with a nucleus, you need to drag that Cas9 to the nucleus. So we often stick on to the protein, we stick on what's called the nuclear localization signal. So that gene is now expressed in the animal and the protein is dragged to the nucleus where the DNA is. And then you add the tracer RNA, which is what gives it the sequence specificity. And this tracer is about 120 bases long, but 20 bases of it are what is called the guide sequence. And the guide sequence binds to DNA, looks for matches, and then when it finds a match, the Cas9 will do the double-stranded break. So all you need to do any targeted cut in any species essentially that they've tested it in is the Cas9 protein and this gRNA. That's it. And like I said, it's been tested. This was back in 2014. The number of successful tests is probably 10 times this now. But every single species they've tried it in, they've gotten it to work. So it's incredibly robust and incredibly simple to use, and this is why it's really taken over the scientific community in a very fast timeframe. So what do you do? I mean, double-stranded breaks, I mean, that doesn't sound particularly interesting. How do you actually use that to create changes in the genome that you want? And I'll talk about all of these techniques, but there are a number of strategies that can be done. The simplest is this one driven by non-homologous end joining. And this is what I use in my lab a lot personally. And that is if you generate a double-stranded break at some frequency, when the cells go to repair it, it can make mistakes. And those mistakes can be small insertions a few bases long or small deletions of a few bases. And when that happens, if you do that, say, in an exon of the gene, you disrupt the reading frame and you can disrupt the gene. So this is how we make knockouts, targeted knockouts in our lab, and we do it literally by the hundreds or thousands. And we do most of our work in zebrafish, and I'll talk about some of these results in zebrafish later on. Another technique is that when you create a double-stranded break, you can do homology-mediated replacement, and I'll talk about how that works. Another technique is you can co-opt the repair machinery by providing a single-stranded DNA template and flood the system with it. And when you create the double-stranded break, when the cells go to repair it, it will use this single-stranded DNA as a template instead of the endogenous DNA, and it'll stick in sequences that weren't supposed to be there. Again, I'll talk about all of these techniques a little more. In a moment. You can do what's called non-homologous end joining of a template, of a target template. You can provide an enormous amount of DNA. You generate a double-stranded break, and by accident, essentially, the huge amount of single-stranded, the double-stranded DNA you put in gets jammed in between the cut. You can do that. Also, you can try to get something to ligate in by creating overlapping cuts and providing a template which has overhangs that get ligated in. So there's a lot of strategies. And like I said, I'll show you how each one of these works in an actual animal. So for zebrafish, and I'll use this as an example just because I know it the best. The way we make a knockout is pretty simple. You create two single-stranded DNAs that overlap some amount of sequence. You add polymerases that extend this to make it about 140 bases long with this T7 promoter on it, and that T7 promoter allows you to synthesize large amounts of RNA. You create these synthetic guide RNAs, sgRNAs, in vitro. You make them in the test tube, then you mix them with either the Cas9 protein or Cas9 mRNA, and these you can inject straight into the single cell of a zebrafish embryo. So these guys are about a millimeter long, so you can do it with a dissecting scope and a micro— in a tiny needle, and you can inject it straight in, and that's usually efficient enough to make a knockout very quickly. In fact, it's so efficient you can do multiple genes simultaneously. This is a paper from a colleague of mine where he actually showed he could knock out 5 genes at varying degrees of efficiency simultaneously. So these are all done in the same fish. And the efficiency range from almost perfect knockouts to about 50% knockout of different genes. So this is a remarkable technique, and we've been using this, like I said, for the last few years to knock out hundreds to thousands of zebrafish genes and test their function. So this is probably the real reason why CRISPR-Cas has taken over the zinc finger and TALEN approaches. And it just works so much better than either of those techniques in terms of targeting efficiencies. We had significant experience with both zinc fingers and TALENs in zebrafish before CRISPRs came out, and our targeting efficiencies were okay. But literally we're seeing 5 to 6 times more targeting activity and the success rate is nearly 100% for CRISPR-Cas. So it's an incredibly exciting technique for us. So if you look at the cumulative data over— this is over about 200 targets that we did. What tends to happen when you just do the targeting and then leave it to biology to replace is you get a lot of deletions. You get a smaller but significant number of insertions. And if you look across all genes, the number of deletions that are in-frame versus frameshift are about basically two-thirds and one-third like you'd predict. Same thing with insertions. So it's essentially what you expect. You sometimes you see very large deletions as well or very large insertions, but most of the knockouts occur plus or minus 10 bases. You can make large deletions in the genome. You just have two sites with CRISPR targets, and you inject them both at the same time, and you can get at a fairly high rate deletions in between the genome. So you can take out whole exons, for example, or very large chunks. The largest I've seen in mouse was over a million bases was deleted from a mouse genome using this kind of approach. Since the original experiments, people have figured out how to do what's called a tissue-specific knockout. Sometimes if you knock a gene out completely, it'll kill the animal, but you're interested in what it does in, say, the liver. And so what you can do is you can express the guide RNA off this plasmid and that's just— this is just cranking out guide RNA at a very high level. And then you can put a promoter in that expresses only in the liver, say. And so this Cas9 is expressed in the liver, and then it meets up with the guide RNA and it creates a knockout, but a knockout that's only happening in the liver and nowhere else. And this works in mice. It works and flies. It's an incredibly powerful approach to generating the kinds of flexible knockouts that you need sometimes to study specific transcription factors or tissue-specific phenomena. All right, so I'm going to talk a little bit about how you make those modifications that are more exotic in the genome after you create a double-stranded break. This was some of the— one of the first experiments published for doing modifications in zebrafish. And as I said, this is what's called the non-homology-based targeted integration. So it turns out if you do a double-stranded break in the genome at a specific locus, that stimulates incorporation of exogenous DNA. So if you provide a very large amount of template in vast excess of the rest of the genome, you put a double-stranded break in at some frequency, and it depends on every— it's highly variable, but at some frequency that DNA gets jammed into the site. And here's proof. In this case, it's jammed into a gene that's expressed in the notochord. And in this case it's being trapped and expressing the red fluorescent protein in the notochord only. And that's proof that that integration occurred at the site they thought it would occur in. A very commonly used one, and this approach is particularly interesting in people that want to correct, say, a genetic mutation that's a single base mistake. And this is using that single-stranded DNA as a template. What's being shown here is an experiment where they're converting blue fluorescent protein into green fluorescent protein, and it's a 3-base change. You can see CCC to CGT, and that, that's enough to change the color of the fluorescent protein. And then they tried different DNA donors to see which one generated the most shift into the green from the blue. And it turns out that down here, this asymmetric template on the strand that isn't being bound by the DNA is the most efficient way to create those kinds of base edits. And in this case, the efficiency was a little over 50%. So a little over 50% of the time, they could write the switch from the template into green fluorescent protein when they used this kind of structure for the target. And 50% is incredibly exciting. That's a really, really good rate of repair. So then the next approach I want to talk about is short-arm homology-directed repair. So in this case, you have a cassette that you want to put in, and in this case it's a GFP, and then you have these homology arms. So these are short sequences, 40 base pairs long, that match the target genome site. So you have a CRISPR that targets your locus, that creates a cut. And then on each side, you have your cassette with the homology arms that match on either side. And this stimulates the correct integration of this vector into the locus. And in here, in this case, they're putting GFP, and it's being expressed in the black spots on the skin. And you can see that GFP is being expressed there. This works quite well, and it also allows you to control the direction of your integration. Other groups have done what we call long-arm homology-directed repair, and in this case, you use much longer arms. The strategy is the same, but the arms are over 1,000 bases in length. The advantage of this is it seems to be that when this goes in, there's no errors. The short-arm homology I talked about, there's often little errors that occur on one side or the other in these homology arms. There's a little bit of little mistakes that are made. But in the case of long-arm homology-directed repair, they don't see those errors. So you can actually swap in very large cassettes into a small area and get tissue-correct expression, gene knockouts, etc., etc. The efficiency is a little bit lower, but it is incredibly precise. So if you need to make a perfect change, this is probably the strategy to take. So since the original Cas9 discovery, probably 90% of the work that's being done with targeted Cas9 is using one Cas from one bacterial species, but there's a lot of work to discover other CasEs with different features that make them more flexible and more powerful. One of the most exciting ones that's been discovered is this one, Cpf1. And the difference between Cpf1 and the Cas9 that's most commonly used is the sequence required, this sequence called a PAM. In Cas9, it's a GG. And in Cpf1 it's a TT. And that may sound like it's not a particularly exciting difference, but it turns out in our genomes and in most vertebrate genomes, there are more As and Ts than there are Cs and Gs. So if you have a TT instead of a GG, there are actually more sites in the genome you can target. And I'll show you the difference here. So this is a stretch of the zebrafish genome. This row here, the black one that says CRISPRs, each one of these black tick marks is a spot in the genome that can be cut by Cas9. And if you look down here, it's the same location. Here's CPF1, and these red tick marks are the sites that can be cut cut by Cpf1. And so you can see there's literally almost 100 times more sites in the genome than there are for the Cas9. So these kinds of changes, these kinds of technology advancements are important and they're happening at a fairly rapid rate these days. Oops, this way. The ability to cut DNA is useful, but it's also useful to be able to bind DNA specifically without causing a genomic disruption, without cutting it. There was a lot of research on the structure of Cas9, and they can see that Cas9 consists of two domains that are both known to be nuclease domains, one Rev-C and HNH. And so each one of these domains is responsible for cutting one of the two strands of the target DNA. So you can actually go through and you can mutate this domain or this domain, and you can make Cas9s that only cut one strand of the DNA. Or in fact, you can mutate both and you can make a Cas9 that cuts no strands of DNA. It just binds specifically. So why would you do something like that? Well, one of the most common things that people are doing with these dead Cas9s, they're called, is that they're linking them to other protein domains and creating transcription factors that they can modulate gene expression without cutting the DNA, right? So You can make a transcriptional repressor. You can take a dead Cas9, you can link it to domains that are known to silence gene expression, and now when this binds close to a gene of interest, you can turn it off, and then probably tissue-specifically, you can turn it off. You can also do transcriptional activation by putting on domains that you know turn gene expression on. And so now you can get ectopic gene expression in tissues or during times when normally that gene isn't expressed. And you can also do modifications to the DNA. So you can do histone modifications, you can do methylation of the DNA, you can link any other kind of modifier and target it to a specific part of the genome. And create disruptions in various ways. And these are all incredibly powerful and interesting techniques that people have been using, using the dead versions of Cas9. So the, as I said, cutting the genome is useful. The techniques I've shown for making modifications are powerful, but they often fail or they are low in efficiency. So people have been looking at other ways to do specific changes in the genome that are maybe safer than creating a double-stranded break. One of these are these new versions of Cas9 that they call base editors. And what these do is it takes the dead Cas9 and now it links it to enzymes that modify bases in the genome. And the first attempt at this was linking Cas9, the dead Cas9, to cytidine deaminase. And what that does is, you know, you bring the dead Cas9 to a specific site in the genome, and you're looking to change a sequence from a C to a T. So cytosine deaminase comes and it binds and it modifies the C. That enzyme changes the C, it actually changes it to a U, and then the DNA repair machinery goes through and says, hey, that's not supposed to be a U, and it fixes it to a T. And then the other— and then there's a mismatch on the other strand, and it has to go through, and you have a 50/50 chance of editing the T back or replacing the G with an A. And so this is a— it's working, it's efficient, and it doesn't actually, at least in some cases, doesn't cause any other damage. It seems to do a site— a a site-specific modification. The trick is often there's more than one C in that same location and you can't necessarily control which C gets modified. So that's sort of the targeting problem that still exists for this kind of strategy. Since that first experiment, there's been a large number of new versions to increase specificity and efficiency of the C-to-T. And then in fact, they've also done an adenosine deaminase, which can change an A to a G. The remarkable thing about the adenosine deaminase is that it doesn't exist in the wild. It was actually a an engineered enzyme modified from existing, other existing enzymes modified to have an activity that doesn't exist in nature. So that was a remarkable engineering feat. So now we have C to T changes and we can do A to G changes. Another strategy is randomization. You can put this AID domain on and then when you target it just does all kinds of random crazy changes And this is a way to do sort of protein evolution and modification of activities in interesting and targeted ways. So I'm going to summarize what I've talked to you about today. So CRISPR-Cas9 really is the preferred technique for most scientists these days. And it has a number of pros and a number of cons. The pros are super simple, very easy to change the targeting and just change RNA, the RNA sequence, which for us is very easy compared to changing a protein sequence, which is a little more difficult. They're very cheap, cost-effective, again, because you don't have to re-engineer it every time. You just change the RNA. It's very inexpensive to use. They're incredibly efficient, much more efficient than some of the other approaches that have been used. They're modular. You can do multiplexing. All of this is incredibly useful and makes this a really amazing technology. There are some cons. The biggest one is off-targets. Sometimes you can have several bases mismatch and it'll still cut and sometimes cut efficiently. If you're— Working in a model organism, that may not be such a big deal, but if you're trying to do gene therapy on someone and you accidentally target another gene, that's a problem. There are still sequence limitations. I told you about the PAM-adjacent motifs. Every CRISPR has some limitations. It needs a few bases, and so you don't have 100% flexibility in the spot that you're targeting. And I'm sure you've probably heard there's intellectual property being argued over. Some very large groups both claim interest in the CRISPR technology, and they can't come to an agreement over who owns the rights. And so that may slow down development in things like gene therapy where there's a lot of money at stake as well as people's health. That's pretty much it I'm talking about today. Here are some links to pretty much all the specific references that I did, that I talked about in today's talk. And I can, I'm sure, provide you with more information if you're interested. Just give me a holler. Belen can provide you with my contact information necessary. And that's it, and thank you.